## Thinking Path > - Paperclip is the open source app people use to manage AI agents for work. > - The runner package already provides the production protocol and execution spine. > - Contributors still need stable SDK surfaces, deterministic test tools, and local inspection tools. > - Those surfaces share generated contracts and must change as one package boundary. > - This pull request adds the package-local SDK, labs, examples, and drift checks. > - The benefit is a reviewable developer platform that does not change application execution selection. ## Linked Issues or Issue Description **Subsystem affected** `packages/paperclip-runner` — runner SDK, conformance tools, and developer tooling. **Problem or motivation** The production runner spine is present, but package consumers cannot build deterministic integrations, inspect sessions, or verify provider-neutral behavior through supported surfaces. **Proposed solution** Add browser, React, standalone, live-session, scenario, conformance, and evaluation surfaces. Add generated contract inventories and package-local verification scripts. Keep production application routing unchanged. **Alternatives considered** We considered splitting each generated catalog, SDK surface, and demo into separate pull requests. Those changes share exports, fixtures, and drift gates. Splitting them would create intermediate package states that do not build. **Roadmap alignment** No overlapping item appears in `ROADMAP.md`. This work extends the runner package that is already on `master`. ## What Changed - Add browser, React, standalone, live-session, and issue-thread SDK surfaces. - Add deterministic mock control-plane, scenario, conformance, replay, and evaluation tools. - Add bounded Codex, OpenCode, and ACPX development transports and fixtures. - Keep deferred managed-provider execution fail-closed. Persisted compatibility data remains readable. - Add generated capability inventories with their source files and drift checks. - Add examples, package documentation, browser checks, and clean-consumer checks. - Preserve the reviewed protocol bounds, replay compatibility aliases, process environment isolation, and semantic redaction limits. - Update the ACPX package patch that the existing workspace patch registry already tracks. - Do not change `pnpm-lock.yaml`, repository workflows, server runtime selection, or the application UI. ## Verification GitHub Actions is the verification authority for this pull request. The repository CI, package TypeScript and Rust checks, package tests, generated-output drift checks, browser checks, security scans, and Greptile review must pass on the exact head. Local test suites were not run because this series uses parallel GitHub Actions for verification. ## Risks This is a large greenfield package change. The main risks are public export drift, generated-output drift, and optional React consumer compatibility. Package boundary checks, clean-consumer checks, and browser tests cover those risks. Production adapter selection and server execution are outside this pull request. ## Stack 1. **This PR:** runner SDK and developer tooling. 2. [Codex production server integration](https://github.com/paperclipai/paperclip/pull/12616). 3. [Provider-neutral task-thread UI](https://github.com/paperclipai/paperclip/pull/12617). ## Model Used OpenAI Codex, GPT-5, high-reasoning mode, with tool use and code execution. ## Checklist - [x] I have included a thinking path that traces from project context to this change - [x] I have specified the model used (with version and capability details) - [x] I have checked ROADMAP.md and confirmed this PR does not duplicate planned core work - [x] I have searched GitHub for duplicate or related PRs and linked them above - [x] I have described the issue in-PR following the feature request template - [x] I have not referenced internal/instance-local Paperclip issues or links - [x] My branch name describes the change and contains no internal Paperclip ticket id - [ ] I have run tests locally and they pass - [x] I have added or updated tests where applicable - [x] I have updated relevant documentation to reflect my changes - [x] I have considered and documented risks above - [ ] All Paperclip CI gates are green - [ ] Greptile is 5/5 with no open P2s, recommendations, or follow-ups - [x] I will address all Greptile and reviewer comments before requesting merge
6.7 KiB
Live console Live Console
Scope
This is the browser layer of Live console. It adds a Live console mode to the
existing standalone devtool in devtools/browser/. It builds on the
protocol and demo server and follows the
approved interaction map and
component decision record.
Every file lives under packages/paperclip-runner/. The console does not
import or change Paperclip server, UI, database, or control-plane code.
Start it
pnpm --filter @paperclipai/paperclip-runner console:live-console
Then open http://127.0.0.1:4180/ and choose Live console.
Live console is deliberately loopback-only. A wildcard or LAN --host fails
startup because this developer console can start provider-backed turns and
resolve approvals. There is no unauthenticated remote mode.
Two environment variables change the driver behind the console:
| Variable | Default | Effect |
|---|---|---|
PAPERCLIP_LIVE_CONSOLE_DRIVER |
demo |
codex starts the real codex app-server driver behind the same routes |
PAPERCLIP_LIVE_CONSOLE_CHUNK_DELAY_MS |
45 |
Milliseconds between streamed chunks in the demo driver |
Where state comes from
The console has one data contract: the canonical PRP event stream and the
SessionSnapshot the shared reducer produces from it. There is no second event
model, no message store, and no client-side session cache.
- The browser reduces the events it receives with the same
applyPrpEventthe server and the CLI use, so a live session and a replayed session project identically. The inspector's Session tab reports that comparison asmatchormismatch. - Transcript order is the reducer timeline order. Nothing is re-sorted.
- Lineage, goal state, pending requests, and the durable cursor come from the demo server's public state, which is refreshed whenever new events arrive.
- The steering chip is the only local UI intent. It resolves to
acknowledged,rejected, orfailedfrom canonical facts and never merges text into the transcript before an acknowledgement.
Surfaces
| Surface | Source of truth | Notes |
|---|---|---|
| Transcript | reducer timeline and items | user messages come from turn.submitted payload text |
| Composer | composer state machine | one visible primary action: Send, Steer, or Stop |
| Steering chip | steering_acknowledgement item or a rejection code |
rejected text stays recoverable with Send as new message |
| Request card | runtime_request.* events |
renders only the actions the upstream request offers |
| Goal banner and menu | capability snapshot and goal events | unsupported verbs are disabled with the exact diagnostic |
| Lineage tree | driver thread lineage | child steering is shown as unsupported, never emulated |
| Connection block | transport status and durable cursor | includes a deliberate connection-loss control |
| Inspector | raw canonical events | four tabs: Events, Requests, Capabilities, Session |
| Replay stepper | recorded canonical events | marked with a REPLAY eyebrow badge |
Interruption races
The console distinguishes all three races the spec requires:
| Race | What the console shows |
|---|---|
| interrupt before start | the pending turn resolves to Cancelled before start; no assistant item is ever created |
| interrupt during generation | the stream stops in place, the partial text stays, and an interrupted divider is added |
| interrupt during a tool call | the tool item keeps its last known status with no fabricated result |
Escape in the composer moves focus to Stop. It never interrupts
directly, because a destructive action needs an explicit activation.
Demo chats
The demo server owns the manifest catalogue and serves it at
GET /api/liveConsole/manifests. Each manifest names the scenario it proves and
the observations a human should confirm; the console renders that list as a
checklist rather than restating it in the UI.
| Manifest | Proves |
|---|---|
completion |
streaming, reasoning, tool, answer, structured result |
steering |
same-turn steering acknowledgement and stale-turn rejection |
interrupt-before-start |
cancellation before the provider accepts the turn |
interrupt-during-generation |
preserved partial text |
interrupt-during-tool |
preserved tool status with no fabricated result |
approvals |
command and file-change approvals, single-response guarantee |
user-input |
typed answers and request expiry |
subagents |
parent/child lineage and unsupported child steering |
goals |
set, view, pause, resume, clear |
goals-unsupported |
exact capability diagnostic instead of a hidden control |
failure |
failed item and failed turn inside the record |
Manifests run in any order. Reset demo state names exactly what it discards and leaves recorded evidence files untouched.
Credentials
No provider or Paperclip credential reaches the browser. The Node process owns the driver, the working directory, and any provider login. Every JSON and event frame passes the demo server's redaction layer, and the inspector renders redaction markers verbatim without attempting to reconstruct them.
The live console and standalone server share one transport admission guard: exact loopback Host/port, loopback sockets, same-origin Origin and Fetch Metadata, plus JSON-only mutations. Browser reset closes the server session; active sessions and SSE subscribers are hard-capped. Local untrusted processes remain a residual risk, and the Codex sandbox/approval policy is the final execution boundary.
Components
Adapted from shadcn/ui and Vercel AI Elements source, rewritten in the package's local idiom, with zero new runtime dependencies:
- AI Elements:
conversation,message,composer,reasoning-item,tool-item. - shadcn/ui:
tabs,menu,dialog,tooltip,banner.
scripts/check-forbidden-imports.mjs fails the build if anything under
devtools/browser/ imports ai, @ai-sdk/*, zod, radix-ui, cmdk,
streamdown, shiki, use-stick-to-bottom, class-variance-authority,
tailwindcss, or nanoid. scripts/check-browser-tokens.mjs fails if a
component file carries a raw colour, pixel, or font value.
Verification
pnpm --filter @paperclipai/paperclip-runner exec vitest run \
src/mock-core/live-console-scripted-driver.test.ts \
src/mock-core/live-console-demo-server.test.ts \
devtools/browser/src/live/transcript-model.test.ts
pnpm --filter @paperclipai/paperclip-runner test:browser
pnpm --filter @paperclipai/paperclip-runner check:browser-tokens
pnpm --filter @paperclipai/paperclip-runner check:forbidden-imports
See the tutorial.